Seasonal variations in the abundance of virus-like particles and bacteria in the floodwater of a Japanese paddy field

Abstract

Viruses are the most abundant biological entities in marine and freshwater environments. Many studies have shown the ecological importance of viruses in the primary production and microbial food web in aquatic environments. However, no studies have examined viral abundance in the floodwater of paddy fields. The present study surveyed the abundance of virus-like particles (VLPs) and bacteria in the floodwater of a Japanese paddy field under a long-term fertilizer trial since 1925 during the rice cultivation period. Virus-like particles and bacterial abundances in the floodwater ranged from 5.6 × 10⁶ to 1.2 × 10⁹ VLPs mL⁻¹ and from 9.2 × 10⁵ to 4.3 × 10⁸ cells mL⁻¹ with mean abundances of 1.5 × 10⁸ VLPs mL⁻¹ and 5.1 × 10⁷ cells mL⁻¹, respectively, and increased with an increase in the turbidity of the floodwater with suspended particles. The magnitude of seasonal variation was more than 50-fold for VLP abundance and 100-fold for bacterial abundance. The virus-to-bacterium ratios fluctuated over the rice cultivation period, ranging from 0.11 to 72 and their increase correlated with the decrease in bacterial abundance. Our results suggest that viral abundance in the floodwater of paddy fields is larger than in natural marine and freshwater environments.

Key words:

• bacteria
• floodwater
• paddy field
• virus
• virus-like particle
• virus-to-bacterium ratio

INTRODUCTION

Viruses are the most abundant biological entities in marine and freshwater environments. And many studies have shown the ecological importance of viruses in global biogeochemical nutrient cycles (Fuhrman 1992; Weinbauer 2004; Weinbauer and Höfle 1998). Viruses are recognized as the greatest genomic reservoirs in marine and freshwater environments because of their huge abundance and diversity (Frost et al. 2005; Paul and Sullivan 2005; Weinbauer and Rassoulzadegan 2004).

The floodwater in paddy fields is a unique, man-made aquatic environment. Several studies have enumerated specific phages in the floodwater using a plate method. Singh (1973) reported that phages of a cyanobacterium Plectonema boryanum strain 594 ranged from zero to 1000 phage forming units (PFU) mL⁻¹ in Cuttack rice fields in India. Ezuka and Kaku (2000) enumerated phages of Xanthomonas oryzae pv. oryzae, a pathogen of bacterial blight in rice plants, in Japanese paddy fields during the nursery and early paddy field stages to enable forecasting of disease severity prior to the appearance of the disease and showed that phages varied from zero to 10² PFU mL⁻¹ according to the disease severity. However, there is no information about viral abundance in the paddy field floodwater, which is an aquatic environment covering 1.6 million ha worldwide (International Rice Research Institute World Rice Statistics: http://www.irri.org/science/ricestat/index.asp).

The type of field management used, for example, the fertilizer application, agrochemical application and drainage, directly influences the floodwater ecosystem in paddy fields. The type of fertilizers, the growth stage of the rice plants and midseason drainage are known to influence the community structure of microorganisms, phytoplankton and zooplankton in the floodwater (Kimura et al. 2002; Okabe et al. 2000; Yamazaki et al. 2001).

The present study aims to enumerate viral and bacterial abundances, and to understand their relationships in the floodwater of a Japanese paddy field during the rice cultivation period. We examined a paddy field under a long-term fertilizer trial to compare the effects of fertilizer type and

Table 1 Total C and N contents, cation exchange capacity and pH (H2O) in the soil

Site	Total C (g kg^−1)	Total N (g kg^−1)	Cation exchange capacity (cmolc kg^−1)	pH
NF	8.5	0.84	8.3	6.1
CF	9.8	1.02	9.0	5.2
CFCa	10.1	1.02	9.0	5.8
CM	23.6	2.07	13.4	5.2
CF, plot with chemical fertilizers; CFCa, plot with chemical fertilizers and lime; CM, plot amended with chemical fertilizers, lime and compost; NF, plot without fertilizer.

the growth stage of the rice plants on viral abundance. This is the first study to examine viral abundance in floodwater, an important aquatic subsystem in the paddy field ecosystem.

MATERIALS AND METHODS

Study site

The study paddy field was located in the Aichi-ken Anjo Research and Extension Center, Central Japan (latitude 34.8° N, longitude 137.5°E). The soil in the field was an Anthraquic Yellow Soil (an Oxyaquic Dystrudept). The field had been subjected to a long-term fertilizer trial since 1925, and the following four plots were chosen for the present study: a plot without fertilizer (NF plot), a plot with chemical fertilizers (CF plot), a plot with chemical fertilizers and lime (CFCa plot), and a plot amended with chemical fertilizers, lime and compost (CM plot). The size of the plots ranged from 3.6 m × 7.3 m to 3.6 m × 9.1 m, with no replication of treatment. The soil properties of the plots are shown in Table 1. Paddy rice (Oryza sativa L. cv. Nihonbare) was cultivated in the plots using conventional means. Irrigation water was introduced from an irrigation canal. The calendar of field management and the amounts of fertilizers and herbicides applied are shown in Table 2.

Sampling of floodwater

Approximately 100 mL of floodwater was collected from the middle part of each of the four sides of each plot, along the footpaths, once every 6–13 days from 15 June 2005 (1 week after transplanting rice) during the cultivation period (11 times). The floodwater was sampled carefully from the upper part to avoid mixing with surface soil soared into the floodwater by disturbance. The depth of floodwater ranged from 1.4 to 15 cm during the cultivation period, and the water requirement in depth was approximately 3 cm day⁻¹ for the experimental plots. In addition, irrigation water was collected on 18 and 24 August and 1, 13 and 22 September. Sampling times were between 10:30 and 11:30 hours. Large aquatic organisms were removed from the water samples by passing the water through a 46-µm sieve. After adding 0.25 m³ m⁻³ glutaraldehyde solution (final concentration 0.005 m³ m⁻³), the water samples were stored in a container with ice. The water samples were kept at −80°C in the laboratory until viral and bacterial enumeration (Marie et al. 1999). Separately, pH and the temperature of the floodwater were determined in the field using a pH meter (SS974, HORIBA Co., Kyoto, Japan).

Enumeration of virus-like particle abundance

Immediately after the preserved floodwater and irrigation water samples were melted at room temperature, they were sonicated for 3 min (38 kHz, 80 W) and vortexed for 30 s, successively. For the enumeration of planktonic viruses (not adsorbed on suspended particles), the floodwater samples were directly filtrated using a 0.45-µm Nuclepore filter (Whatman Japan KK, Tokyo, Japan), diluted 10-fold to 100-fold with TE buffer (10 mmol L⁻¹ Tris, 1 mmol L⁻¹ ethylenediaminetetraacetic acid [pH 8]; Brussaard et al. 2000), and stained with SYBR Green I (final concentration 0.0001 m³ m⁻³; Molecular Probes, Eugene, OR, USA) for 15 min at 80°C under dark conditions (Brussaard et al. 2000). SYBR Green I is the most stable fluorescent dye used for viral enumeration (Wommack and Colwell 2000). Enumeration of planktonic viruses was conducted for the samples collected from the right and left side footpaths of the irrigation inlet footpath on 15 June, 7 July, 2 August, 18 August and 13 September.

For the enumeration of total viral abundance, 1 mL of the water samples was placed into a 20-mL glass bottle

Table 2 Calendar of field management in the study paddy field in 2005

Flooding	7 June	First topdressing^§	29 July
Puddling	8 June	Second top dressing^§	8 August
Basal fertilization^†	8 June	Drainage	28 September
Transplanting	9 June	Harvesting	26 October
Herbicides^‡	20 June
Midseason drainage	15–25 July
^†60 kg N, 86 kg P2O5 and 56 kg K2O in the CF plot, 60 kg N, 86 kg P2O5, 56 kg K2O and 1.1 t CaO in the CFCa plot and 60 kg N, 86 kg P2O5 and 56 kg K2O, 1.1 t CaO and 22.5 t of rice straw compost in the CM plot. N, P, K and Ca were applied as (NH4)2SO4, calcium superphosphate, KCl and lime, respectively.
^‡1.5 kg bentiocarb, 51 g bensulfuron-methyl and 300 g ha^−1 mefenacet.
^§20 kg N ha^−1 in the plots with fertilization. CF, plot with chemical fertilizers; CFCa, plot with chemical fertilizers and lime; CM, plot amended with chemical fertilizers, lime and compost.

after the ultrasonication and vortexing described previously, and 1 mL of 36 g L⁻¹ nutrient broth (NB) solution (Eiken Chemical Co., Tokyo, Japan) was added, and again sonicated for 3 min and vortexed for 30 s (Danovaro et al. 2001). The NB solution was used to dislodge viruses from soil particles (Germida and Casida 1981; Lanning and Williams 1982). The viruses were then separated from the suspended particles by passing through a 0.45-µm Nuclepore filter. The filtrate was diluted 10-fold to 100-fold with TE buffer, and stained with SYBR Green I.

Enumeration of viruses was carried out using epifluorescence microscopy (EFM). Stained viruses were collected on a 0.022-µm pore-size Anodisc 25 membrane filter (25 mm diameter; Whatman Japan KK) that had been stained beforehand with Irgaran Black for absorbing autofluorescence of the filter, placed on a glass slide, and guarded from color fading with 20 µL of SlowFade Gold Antifade reagent (Invitrogen Japan KK, Tokyo, Japan). Viruses on the filters were observed at a magnification of ×1,000 using an Olympus BX-50 epifluorescence microscope equipped with a UPlan Fl Objective lens with U-MNIBA filter (Olympus, Tokyo, Japan), and 6–8 fields of the scope were randomly recorded using a digital camera (shutter speed 10 s). Large particles were regarded as bacterial particles, and viruses were enumerated by counting the smaller particles in 2–4 filters for every sample. As all the viral particles that were counted by EFM were not confirmed to be viruses, we called the enumerated viruses virus-like particles (VLPs) in the present study (Wommack and Colwell 2000).

To check the reliability of viral enumeration using EFM, a portion of the filtrate that passed through a 0.45-µm Nuclepore filter was subjected to observation using a transmission electron microscope (TEM; H-7500AMT Advantage HR, Hitachi High-Technologies, Tokyo, Japan). Floodwater samples collected from every plot on 15 and 28 June and 7 July were used. Viruses in 0.5 mL of the filtrate were collected onto formvar-coated 400 mesh Cu grids using ultracentrifugation (100,000 g for 60 min; Beckman TL-100 tabletop ultracentrifuge, Beckman Coulter KK, Tokyo, Japan) according to the method of Børsheim et al. (1990). The TEM enumeration was carried out by taking photographs of more than 20 fields after negative staining with 20 g L⁻¹ uranyl acetate.

The sample bottles in the VLP enumeration were glass to prevent the viruses adhering to the surface of the bottles (Thompson et al. 1998). Distilled water was used for the preparation of the NB solution and the TE buffer, and the SYBR Green I solution was rendered virus-free by passing it through a 0.022-µm filter.

Enumeration of bacterial abundance

The melted water samples were ultrasonicated and vortexed as described above. One milliliter of the sample was placed into a 15-mL centrifuge tube (Asahi Technoglass Co., Tokyo, Japan) and mixed with 1 mL of TE buffer. Bacterial staining was carried out by the addition of SYBR Green I reagent (final concentration 0.0001 m³ m⁻³) and subsequent incubation for 30 min at room temperature under dark conditions. Stained bacteria were collected on a 0.22-µm pore-size polycarbonate 25 black membrane filter (25 mm diameter; Advantec Toyo Kaisha, Tokyo, Japan). The membrane filter was placed on a glass slide, and 20 µL of SlowFade Gold Antifade reagent was added. Bacterial abundance was counted for 7–13 fields measuring 50 µm × 50 µm using an Olympus BX-50 epifluorescence microscope equipped with a UPlan Fl Objective lens with U-MNIBA filter.

Measurement of the turbidity of floodwater samples

Immediately after the melted water samples were ultrasonicated and vortexed as described above, the turbidity of the floodwater samples was determined using a spectrophotometer (UV-2200, Shimadzu Co., Kyoto, Japan) with a wavelength of 660 nm. In addition, the dry weight of the suspended particles in sample water collected on 28 June and 7 July was determined after desalting the floodwater by centrifuging the water twice at 8,500 g in distilled water, and the absorption coefficient (absorbance per dry weight) was calculated. The dry weight was determined after drying the samples at 105°C.

Statistical analyses

A correlation matrix was calculated to estimate the correlation among environmental properties, VLP abundance, bacterial abundance, and virus-to-bacterium ratios (VBRs). To identify the sampling date and the plot that was significantly different from the other samples for the respective variables, two-way anovas and Tukey's honestly significant difference tests were carried out (SPSS for Windows, Ver. 10.0). In addition, as midseason drainage is known to affect the communities of microorganisms, phytoplankton and zooplankton in paddy fields (Sugano et al. 2005; Tanahashi et al. 2005; Yamazaki et al. 2001, 2003), the influences of the midseason drainage during cultivation (15–25 July) were also statistically analyzed using a Tukey's honestly significant difference test.

RESULTS AND DISCUSSION

Temperature, pH and turbidity of the floodwater

Floodwater temperature ranged from 20.6 to 35.3°C during the rice cultivation period (Fig. 1a). The highest temperature was recorded on 2 August in the CF, CFCa and CM plots, and on 13 September in the NF plot.

Figure 1  Seasonal variations in (a) temperature, (b) pH and (c) turbidity determined by absorbance at 660 nm wavelength in the floodwater samples in a plot without fertilizer (○), a plot with chemical fertilizers (□), a plot with chemical fertilizers and lime (⧫) and a plot amended with chemical fertilizers, lime and compost (▴) during the rice cultivation period. Shaded columns show the period of midseason drainage. Values are means ± standard deviation; n = 4.

Floodwater temperature was higher in the NF plot than the other plots after midseason drainage (P < 0.01), probably because poor growth of rice plants allowed efficient light penetration into the plot (the total biomass of the aboveground part at harvest time was 3.5, 9.7, 11.8 and 17.1 × 10³ kg ha⁻¹ for the NF, CF, CFCa and CM plots, respectively).

Floodwater pH ranged from 5.9 to 10.0 during the rice cultivation period (Fig. 1b). Floodwater pH was significantly correlated with floodwater temperature throughout the cultivation (P < 0.01). After the midsummer drainage, the NF plot showed a higher floodwater pH than the other plots (Fig. 1b: P < 0.01), in part, because of the active photosynthesis of phytoplankton resulting in a decrease in HCO⁻ ₃ concentration (the increase in OH⁻; HCO⁻ ₃  = CO₂[to photosynthesis] + OH⁻) in the floodwater (Mikkelsen et al. 1978). Higher pH was also observed after midseason drainage in the CFCa plot compared with the CF and CM plots (P < 0.01), which was attributed to the application of lime. In general, variations in pH among the sampling sites were relatively small except for the CF plot on 18 August and 13 September, the CFCa plot on 10 August, and the CM plot on 18 August, when extremely high pHs (8.9 or higher) were observed at one of the four sampling sites. As the pH on 15 June was higher than on 28 June (P < 0.01), the effect of herbicide application (20 June) on the floodwater pH was estimated (through the suppression of photosynthesis by phytoplankton; Fig. 1c).

Turbidity of the floodwater estimated from the absorbance at 660 nm was higher before the midseason drainage than after it in every plot (Fig. 1c; P < 0.01). In addition, turbidity in the CF plot (mean ± standard deviation [SD]: 0.023 ± 0.017) and the CM plot (0.025 ± 0.015) was larger than in the NF plot (0.012 ± 0.006) and the CFCa plot (0.011 ± 0.007) after the midseason drainage (P < 0.05), although these values were low. Turbidity correlated well with floodwater pH throughout the flooded period (P < 0.05) and with floodwater temperature after the midseason drainage (P < 0.01). Suspended substances were estimated to consist mainly of soil particles from the brown color and texture of the suspended substances collected on the Nuclepore filters. The higher floodwater pH seemed to have accelerated the dispersion of soil particles by ionizing the sites of variable charges on soil particles negatively.

Validity of viral enumeration using EFM

Estimation of viral abundance using TEM is the standard method. However, enumeration of viruses using EFM is now more frequently used than TEM to estimate total viral abundance in aquatic environments because of the rapidity and lower costs of the former method (Weinbauer 2004; Wommack and Colwell 2000). As

Figure 2  Comparison of viral counts using epifluorescence microscopy (EFM) and transmission electron microscopy (TEM). The line indicates the linear regression (EFM = 1.27 × TEM −13.80, R2 = 0.824).

this is the first study to enumerate viruses in the floodwater of paddy fields, we evaluated the reliability of the EFM method used in the present study by comparing the viral abundance determined using EFM and TEM.

As shown in Fig. 2, VLP abundance estimated using EFM showed a highly significant correlation (P < 0.01) with the abundance determined using TEM, on average 33% larger viral counts were recorded using the former method. Overestimation of viral abundance using EFM or underestimation using TEM is a common occurrence in viral enumeration of aquatic samples (Weinbauer 2004; Wommack and Colwell 2000), and Noble and Fuhrman (1998) also found on average 28% larger viral counts using EFM with SYBR Green I for coastal water samples. The overestimation using EFM resulted, in part, from the counting of dots of non-viral origins. Underestimation using TEM was attributable to viral loss when uranyl acetate was wicked away from the grids. Furthermore, viruses might be obscured by other larger, darkly stained particles on the grids (Noble and Fuhrman 1998). Non-viral origins in the present study may be attributed to bacteria that could pass through the 0.45-µm filter. However, VLPs in the floodwater were generally larger than the bacterial abundance and the large fluorescent particles of bacterial sizes could be excluded visually, although the bacterial portion that passed through a 0.45-µm filter is generally greater than 50% (Cole et al. 1993). Therefore, the potential bias exerted by incorrect counting of bacteria as viruses would not be serious in the present study. On the whole, we regarded our method of VLP enumeration as satisfactory for the purposes of the present study.

Abundance of virus-like particles

The VLP abundance in the floodwater during the rice cultivation period ranged from 1.3 × 10⁷ to 6.4 × 10⁸ (mean ± SD: 1.5 ± 1.4 × 10⁸) VLPs mL⁻¹ in the NF plot, from 5.6 × 10⁶ to 8.1 × 10⁸ (1.6 ± 2.1 × 10⁸) VLPs mL⁻¹ in the CF plot, from 6.1 × 10⁶ to 1.2 × 10⁹ (1.8 ± 3.2 × 10⁸) VLPs mL⁻¹ in the CFCa plot, and from 8.6 × 10⁶ to 9.4 × 10⁸ (1.0 ± 1.7 × 10⁸) VLPs mL⁻¹ in the CM plot (Fig. 3a). Thus, VLP abundance fluctuated by 50 to 200-fold. It tended to increase with time after transplanting first until the beginning of July and VLPs were more numerous on 28 June and 7 July than at the other sampling times (P < 0.01 except for P < 0.02 between 15 June and 7 July). The largest and the smallest fluctuations in VLP abundance during the rice cultivation period were observed in the CFCa and NF plots, respectively. The VLPs were more numerous in the NF plot than in the CFCa and CM plots after midseason drainage (P < 0.01). The VLP abundance in the irrigation water ranged from 5.1 × 10⁶ to 1.1 × 10⁷ VLPs mL⁻¹ (Fig. 3a), which was smaller than that in the floodwater. No correlation was found between the VLP abundances.

Figure 3  Seasonal variations in the abundance of (a) virus-like particles (VLPs) and (b) bacteria in the water samples in a plot without fertilizer (○), a plot with chemical fertilizers (□), a plot with chemical fertilizers and lime (⧫), a plot amended with chemical fertilizers, lime and compost (▴), and the irrigation water (★) during the rice cultivation period. Shaded columns show the period of midseason drainage. Values are means ± standard deviation; n = 4.

Figure 4  Relationships between (a) virus-like particles (VLPs) and (b) bacterial abundance and floodwater turbidity in a plot without fertilizer (○), a plot with chemical fertilizers (□), a plot with chemical fertilizers and lime (⧫) and a plot amended with chemical fertilizers, lime and compost (▴) during the rice cultivation period. Correlation coefficients were 0.688 (significant at the 1% level) between the abundance of VLPs and the floodwater turbidity and 0.691 (significant at the 1% level) between the abundance of bacteria and the floodwater turbidity.

These findings on VLP abundance in the floodwater suggest that the floodwater is an aquatic environment abundant in viruses compared with marine, estuarine and freshwater systems, where VLP abundance is generally less than 10⁷ VLPs mL⁻¹ (Weinbauer 2004; Wommack and Colwell 2000). The floodwater is a very shallow aquatic environment (15 cm or less for the study paddy field), and the nutrient supply from the soil and fertilizers might result in active growth of microorganisms and a subsequent increase in viruses. The floodwater was also characterized by large seasonal variations of more than 50 to 200-fold in VLP abundance. Seasonal variation of viruses is mainly attributed to the autumn and spring blooms of phytoplankton in natural aquatic environments, but the magnitude of variation was less than 30-fold at most (Bratbak et al. 1990; Larsen et al. 2001; Wilson et al. 2002). Smaller VLP abundance in the irrigation water compared with the floodwater indicated that viruses in the floodwater were not derived directly from the irrigation water, but rather occurred in the floodwater at least after the midseason drainage (Fig. 3a).

The VLP abundance in every plot correlated significantly with turbidity (Fig. 4a; P < 0.01). Hewson et al. (2001) also found correlations between VLP abundance and suspended particles in the water columns of eutrophic and oligotrophic estuaries in subtropical Queensland, Australia, and they attributed the benthic viral populations as the settlement of planktonic viruses by adsorbing to suspended particles (> 0.7 µm) in the water column, in which suspended particles mainly consisted of organic matter.

The VLP abundance also correlated well with floodwater pH after midseason drainage in the CFCa plot (P < 0.01). Significant relationships in VLP abundance with floodwater temperature were only observed in the CF plot (P < 0.01) and the CFCa plot (P < 0.05) after midseason drainage, as well as in the CM plot during the flooding period (P < 0.05).

Abundance of planktonic virus-like particles

The VLP abundance of the planktonic forms ranged from 2.2 × 10⁶ to 8.0 × 10⁷ VLPs mL⁻¹ (mean ± SD: 2.0 ± 1.7 × 10⁷ VLPs mL⁻¹), with greater abundance in the NF plot compared with the CF and CFCa plots (P < 0.05). The abundance was not significantly different among the sampling times, and accounted for 1.7 to 106% of the total VLP abundance in the floodwater (Fig. 5). The proportion of planktonic forms in the total viral populations was small when VLP abundance was numerous, and the average proportions were 8.1, 5.8, 58, 40 and 46% on 15 June, 7 July, 2 August, 18 August and 13 September, respectively.

Abundance of VLPs adsorbed to suspended particles in the sample water (VA; ×10⁷ VLPs mL⁻¹) was estimated by subtracting the abundance of planktonic VLPs from the abundance of the total VLPs (Figs. 3a,5). The VA significantly correlated with the weights of the suspended particles (x; g L⁻¹); VA = 560x − 2.7 (r = 0.7255, P < 0.01), where the absorption coefficient (absorbance 1 = 0.71 mg mL⁻¹) was used to translate the turbidity of floodwater samples to the weight of their suspended particles. The average proportions of particle-adsorbed forms of VLPs in the floodwater were 75% during the rice cultivation period. Larger VLP abundance and seasonal variation in the floodwater in comparison with other aquatic environments were, therefore, estimated to be because of the presence of suspended particles in the

Figure 5  Percentage of planktonic virus-like particles (VLPs) in the total viral populations. The figure shows the average values with a half difference between duplication. (□), 15 June; (▓), 7 July; (▒), 2 August; (░), 18 August; (▀), 13 September. CF, plot with chemical fertilizers; CFCa, plot with chemical fertilizers and lime; CM, plot amended with chemical fertilizers, lime and compost; NF, plot without fertilizer.

floodwater. Field management, such as puddling and transplanting, might have resulted in soil dispersion. In addition, drying the surface soil by midseason drainage (from 15 July to 25 July) might have decreased VLP abundance in the floodwater after the midseason drainage, because midseason drainage deposited suspended particles and prevented them from being dispersed again into the floodwater by consolidation of the surface soil (Fig. 1c). Thus, field management might have totally resulted in the large variation in VLP abundance in the floodwater.

Bacterial abundance

As shown in Fig. 3b, bacterial abundance ranged from 3.0 × 10⁶ to 3.4 × 10⁸ (mean ± SD: 4.7 ± 6.2 × 10⁷) cells mL⁻¹ in the NF plot, from 6.9 × 10⁵ to 3.9 × 10⁸ (6.1 ± 10.0 × 10⁷) cells mL⁻¹ in the CF plot, from 1.0 × 10⁶ to 3.9 × 10⁸ (4.7 ± 9.1 × 10⁷) cells mL⁻¹ in the CFCa plot, and from 9.2 × 10⁵ to 4.3 × 10⁸ (4.7 ± 8.9 × 10⁷) cells mL⁻¹ in the CM plot. Bacterial abundance fluctuated by 100 to 600-fold over the rice cultivation period (Fig. 3b). Bacterial abundance in the floodwater on 28 June and 7 July was more numerous than at the other sampling times (P < 0.01), just as VLP abundance was. The largest and smallest fluctuations in bacterial abundance over the rice cultivation period were observed in the CF and NF plots, respectively. There was no significant difference in bacterial abundance among the plots after the midseason drainage. Bacterial abundance in the irrigation water ranged from 8.4 × 10⁵ to 6.2 × 10⁷ cells mL⁻¹ (Fig. 3b), smaller than the abundance in the floodwater, and did not correlate with that in the floodwater. These facts indicated that bacteria in the floodwater were not derived from the irrigation water, at least after the midseason drainage, as in the case of viruses.

Bacterial abundance in every plot was also significantly correlated with turbidity (P < 0.01; Fig. 4b). Bacterial abundance in the floodwater samples on 28 June and 7 July ranged from 1.29 × 10⁸ to 3.22 × 10⁸ cells mL⁻¹, and suspended particles ranged from 0.05 to 0.22 mg mL⁻¹ for those samples. Assuming that all the bacterial cells in the floodwater are coccoid with the diameter of 1 µm and contain 90% water, their dry weight amounted to 0.0068–0.0169 mg mL⁻¹, and bacteria were estimated to be 4–14% of suspended particles on the weight basis in those floodwater samples. As the average diameter of bacteria in lakes is far smaller than 1 µm (Cole et al. 1993), the contribution of bacteria to suspended particles might be less than 10%. Therefore, suspended particles were estimated to consist mainly of soil particles as estimated before from the appearance of them. Suspended particles contribute to an increase not only in VLP abundance but also in bacterial abundance in the floodwater. Most of the bacteria under a microscope adsorbed to suspended particles in the present study (data not shown). Compared with VLP, bacterial abundance is more sensitive to fluctuations in turbidity.

Virus-to-bacterium ratios

As shown in Fig. 6, VBRs ranged from 0.81 to 23 (mean ± SD: 7.1 ± 7.3) in the NF plot, from 0.40 to 72 (7.9 ± 11.5) in the CF plot, from 0.80 to 59 (10.2 ± 12.0) in the CFCa plot, and from 0.11 to 47 (7.4 ± 10.2) in the CM plot. The VBRs in the irrigation water ranged from 0.17 to 6.1. There was no significant difference in VBRs among the plots. As expected from Fig. 3, VLP abundance showed a significant correlation with bacterial abundance for every plot (P < 0.01). The VBRs tended to be higher in August after the midseason drainage (11.5 ± 13.9 to 17.4 ± 11.3) than in June and July before the midseason drainage (3.8 ± 2.7 to 4.3 ± 4.8) and in September (1.5 ± 1.2 to 8.3 ± 7.3). The higher VBRs in August resulted, in part, from the lower abundance of bacterial communities in August compared with those in the other months.

In contrast to the general findings that VBRs fall between 3 and 25 in aquatic environments (Fuhrman 1999; Weinbauer 2004; Wommack and Colwell 2000), 35% and 6% of the floodwater samples in this study showed lower and higher VBRs, respectively, than the range recorded in aquatic environments (Fig. 6). Weinbauer (2004) reviewed the results on VBRs in marine sediments and found that they ranged from 0.11 to 71, which may indicate that shallow aquatic environments

Figure 6  Seasonal variations in virus-to-bacterium ratios (VBRs) in the floodwater in a plot without fertilizer (○), a plot with chemical fertilizers (□), a plot with chemical fertilizers and lime (◊) and a plot amended with chemical fertilizers, lime and compost (▴) during the rice cultivation period. The correlation coefficient between the abundance of VLPs and bacteria was 0.608 (significant at the 1% level).

Figure 7  Relationships between virus-like particles (VLPs) and bacterial abundance and virus-to-bacterium ratios (VBRs) in the floodwater in a plot without fertilizer (○), a plot with chemical fertilizers (□), a plot with chemical fertilizers and lime (⧫) and a plot amended with chemical fertilizers, lime and compost (▴) during the rice cultivation period. The correlation coefficients were 0.041 (not significant) between VLPs and VBRs and −0.296 (significant at the 1% level) between bacteria and VBRs.

such as the floodwater that are strongly affected by sediments underneath fluctuate VBRs widely. The VBRs may be regulated by host–virus relationships in terms of nutritional and environmental conditions in general aquatic environments. In contrast, most of the viruses and bacteria were located on suspended particles in the floodwater, when viruses were numerous (Figs 3,4,5), and those particles might have originated from the soil surface. Therefore, the site of proliferation of viruses and bacteria of particle-adsorbed forms was not clear, and adsorbing properties of viruses and bacteria to those particles should be considered in order to understand the change in VBRs in the floodwater. Greater changes in bacterial abundance with changes in turbidity (Fig. 4b) indicate that a higher proportion of bacteria exist as adsorbed forms than viruses.

There was no significant correlation between VLP abundance and VBRs in total, although there were positive correlations between them when the correlation was calculated separately among the floodwater samples before (P < 0.01) and after (P < 0.05) the midseason drainage (Fig. 7a). Seasonal variation in VBRs showed significant correlation with the change in bacterial abundance, and VBRs increased with the decrease in bacterial abundance in the floodwater (Fig. 7b; P < 0.01). The inverse relationship between VBRs and bacterial abundance observed in the present study is common in aquatic environments (Wommack and Colwell 2000).

In conclusion, the floodwater of paddy fields is a unique aquatic environment for viruses and bacteria with their higher abundances compared with other aquatic environments. The floodwater was characterized by large seasonal variations in viral and bacterial abundances with magnitudes of more than 50-fold for VLP abundance and 100-fold for bacterial abundance, and the largest and smallest fluctuations in VLP abundance during the rice cultivation period were observed in the CFCa and NF plots, respectively. Viruses and bacteria in the floodwater were not directly derived from the irrigation water, but rather occurred in the floodwater. The VLP and bacterial abundances correlated with the turbidity (the amount of suspended particles) of the floodwater, and suspended particles were estimated to consist mainly of soil particles. The VBRs fluctuated widely during rice cultivation, and were negatively correlated with changes in bacterial abundance.

ACKNOWLEDGMENTS

The authors thank Professor Hiroshi Miyake and Assistant Professor Michio Kawasaki, Graduate School of Bioagricultural Sciences, Nagoya University, for their technical assistance with the transmission electron microscope manipulation. This study was conducted with the financial support of the Sumitomo Foundation for Environmental Sciences in 2005.